Online Goal Babbling for rapid bootstrapping of inverse models in high dimensions

We present a model of online Goal Babbling for the bootstrapping of sensorimotor coordination. By modeling infants' early goal-directed movements we show that inverse models can be bootstrapped within a few hundred movements even in very high-dimensional sensorimotor spaces. Our model thereby explains how infants might initially acquire reaching skills without the need for exhaustive exploration, and how robots can do so in a feasible way. We show that online learning in a closed loop with exploration allows substantial speed-ups and, in high-dimensional systems, outperforms previously published methods by orders of magnitude.